End of training

all_results.json

@@ -1,8 +1,8 @@
 {
     "epoch": 3.25,
-    "eval_accuracy": 0.7043010752688172,
-    "eval_loss": 0.8327615261077881,
-    "eval_runtime": 1683.6481,
-    "eval_samples_per_second": 0.442,
-    "eval_steps_per_second": 0.055
+    "eval_accuracy": 0.7163978494623656,
+    "eval_loss": 0.7914324402809143,
+    "eval_runtime": 1837.684,
+    "eval_samples_per_second": 0.405,
+    "eval_steps_per_second": 0.051
 }

test_results.json

@@ -1,8 +1,8 @@
 {
     "epoch": 3.25,
-    "eval_accuracy": 0.7043010752688172,
-    "eval_loss": 0.8327615261077881,
-    "eval_runtime": 1683.6481,
-    "eval_samples_per_second": 0.442,
-    "eval_steps_per_second": 0.055
+    "eval_accuracy": 0.7163978494623656,
+    "eval_loss": 0.7914324402809143,
+    "eval_runtime": 1837.684,
+    "eval_samples_per_second": 0.405,
+    "eval_steps_per_second": 0.051
 }

trainer_state.json

@@ -1,5 +1,5 @@
 {
-  "best_metric": 0.7328859060402685,
+  "best_metric": 0.72751677852349,
   "best_model_checkpoint": "vivit-b-16x2-kinetics400-finetuned-crema-d/checkpoint-2976",
   "epoch": 3.248991935483871,
   "eval_steps": 500,
@@ -10,2144 +10,2144 @@
   "log_history": [
     {
       "epoch": 0.0,
-      "grad_norm": 10.041251182556152,
+      "grad_norm": 8.909350395202637,
       "learning_rate": 1.6778523489932886e-06,
-      "loss": 2.0618,
+      "loss": 1.8749,
       "step": 10
     },
     {
       "epoch": 0.01,
-      "grad_norm": 12.547879219055176,
+      "grad_norm": 10.488879203796387,
       "learning_rate": 3.3557046979865773e-06,
-      "loss": 1.9746,
+      "loss": 1.8028,
       "step": 20
     },
     {
       "epoch": 0.01,
-      "grad_norm": 9.306560516357422,
+      "grad_norm": 9.420026779174805,
       "learning_rate": 5.033557046979865e-06,
-      "loss": 1.8185,
+      "loss": 1.7855,
       "step": 30
     },
     {
       "epoch": 0.01,
-      "grad_norm": 7.815144062042236,
+      "grad_norm": 8.927688598632812,
       "learning_rate": 6.7114093959731546e-06,
-      "loss": 1.8203,
+      "loss": 1.8411,
       "step": 40
     },
     {
       "epoch": 0.02,
-      "grad_norm": 10.477940559387207,
+      "grad_norm": 10.278192520141602,
       "learning_rate": 8.389261744966444e-06,
-      "loss": 1.8039,
+      "loss": 1.7949,
       "step": 50
     },
     {
       "epoch": 0.02,
-      "grad_norm": 8.257948875427246,
+      "grad_norm": 8.889835357666016,
       "learning_rate": 1.006711409395973e-05,
-      "loss": 1.7131,
+      "loss": 1.7569,
       "step": 60
     },
     {
       "epoch": 0.02,
-      "grad_norm": 7.9578776359558105,
+      "grad_norm": 8.13683795928955,
       "learning_rate": 1.174496644295302e-05,
-      "loss": 1.701,
+      "loss": 1.677,
       "step": 70
     },
     {
       "epoch": 0.03,
-      "grad_norm": 11.580317497253418,
+      "grad_norm": 12.050573348999023,
       "learning_rate": 1.3422818791946309e-05,
-      "loss": 1.5872,
+      "loss": 1.6716,
       "step": 80
     },
     {
       "epoch": 0.03,
-      "grad_norm": 12.274490356445312,
+      "grad_norm": 12.406548500061035,
       "learning_rate": 1.51006711409396e-05,
-      "loss": 1.5683,
+      "loss": 1.6021,
       "step": 90
     },
     {
       "epoch": 0.03,
-      "grad_norm": 9.367303848266602,
+      "grad_norm": 9.213874816894531,
       "learning_rate": 1.6778523489932888e-05,
-      "loss": 1.557,
+      "loss": 1.5515,
       "step": 100
     },
     {
       "epoch": 0.04,
-      "grad_norm": 7.54719877243042,
+      "grad_norm": 7.81805419921875,
       "learning_rate": 1.8456375838926178e-05,
-      "loss": 1.4923,
+      "loss": 1.4358,
       "step": 110
     },
     {
       "epoch": 0.04,
-      "grad_norm": 8.119046211242676,
+      "grad_norm": 7.475973606109619,
       "learning_rate": 2.013422818791946e-05,
-      "loss": 1.4376,
+      "loss": 1.4428,
       "step": 120
     },
     {
       "epoch": 0.04,
-      "grad_norm": 8.935946464538574,
+      "grad_norm": 8.327278137207031,
       "learning_rate": 2.181208053691275e-05,
-      "loss": 1.5429,
+      "loss": 1.5263,
       "step": 130
     },
     {
       "epoch": 0.05,
-      "grad_norm": 7.0674147605896,
+      "grad_norm": 8.09311580657959,
       "learning_rate": 2.348993288590604e-05,
-      "loss": 1.3362,
+      "loss": 1.3745,
       "step": 140
     },
     {
       "epoch": 0.05,
-      "grad_norm": 11.671470642089844,
+      "grad_norm": 11.201431274414062,
       "learning_rate": 2.516778523489933e-05,
-      "loss": 1.371,
+      "loss": 1.2682,
       "step": 150
     },
     {
       "epoch": 0.05,
-      "grad_norm": 11.36544132232666,
+      "grad_norm": 10.589693069458008,
       "learning_rate": 2.6845637583892618e-05,
-      "loss": 1.3582,
+      "loss": 1.2924,
       "step": 160
     },
     {
       "epoch": 0.06,
-      "grad_norm": 11.540151596069336,
+      "grad_norm": 12.334148406982422,
       "learning_rate": 2.8523489932885905e-05,
-      "loss": 1.3666,
+      "loss": 1.3283,
       "step": 170
     },
     {
       "epoch": 0.06,
-      "grad_norm": 12.664009094238281,
+      "grad_norm": 11.779102325439453,
       "learning_rate": 3.02013422818792e-05,
-      "loss": 1.5347,
+      "loss": 1.5154,
       "step": 180
     },
     {
       "epoch": 0.06,
-      "grad_norm": 8.551016807556152,
+      "grad_norm": 7.225616455078125,
       "learning_rate": 3.1879194630872485e-05,
-      "loss": 1.3923,
+      "loss": 1.3195,
       "step": 190
     },
     {
       "epoch": 0.07,
-      "grad_norm": 10.388668060302734,
+      "grad_norm": 10.551006317138672,
       "learning_rate": 3.3557046979865775e-05,
-      "loss": 1.3822,
+      "loss": 1.3953,
       "step": 200
     },
     {
       "epoch": 0.07,
-      "grad_norm": 10.217473983764648,
+      "grad_norm": 9.995854377746582,
       "learning_rate": 3.523489932885906e-05,
-      "loss": 1.2387,
+      "loss": 1.1524,
       "step": 210
     },
     {
       "epoch": 0.07,
-      "grad_norm": 10.830105781555176,
+      "grad_norm": 8.858394622802734,
       "learning_rate": 3.6912751677852356e-05,
-      "loss": 1.2035,
+      "loss": 1.1582,
       "step": 220
     },
     {
       "epoch": 0.08,
-      "grad_norm": 14.143545150756836,
+      "grad_norm": 14.140151977539062,
       "learning_rate": 3.859060402684564e-05,
-      "loss": 1.2766,
+      "loss": 1.2541,
       "step": 230
     },
     {
       "epoch": 0.08,
-      "grad_norm": 12.051302909851074,
+      "grad_norm": 12.274311065673828,
       "learning_rate": 4.026845637583892e-05,
-      "loss": 1.2969,
+      "loss": 1.2666,
       "step": 240
     },
     {
       "epoch": 0.08,
-      "grad_norm": 7.3714823722839355,
+      "grad_norm": 9.257830619812012,
       "learning_rate": 4.194630872483222e-05,
-      "loss": 1.1517,
+      "loss": 1.1348,
       "step": 250
     },
     {
       "epoch": 0.09,
-      "grad_norm": 9.341689109802246,
+      "grad_norm": 8.932195663452148,
       "learning_rate": 4.36241610738255e-05,
-      "loss": 1.2486,
+      "loss": 1.2301,
       "step": 260
     },
     {
       "epoch": 0.09,
-      "grad_norm": 9.874135971069336,
+      "grad_norm": 8.28331470489502,
       "learning_rate": 4.530201342281879e-05,
-      "loss": 1.1304,
+      "loss": 1.15,
       "step": 270
     },
     {
       "epoch": 0.09,
-      "grad_norm": 10.551310539245605,
+      "grad_norm": 14.461810111999512,
       "learning_rate": 4.697986577181208e-05,
-      "loss": 1.1465,
+      "loss": 1.1571,
       "step": 280
     },
     {
       "epoch": 0.1,
-      "grad_norm": 14.167583465576172,
+      "grad_norm": 17.485107421875,
       "learning_rate": 4.865771812080537e-05,
-      "loss": 1.2225,
+      "loss": 1.0857,
       "step": 290
     },
     {
       "epoch": 0.1,
-      "grad_norm": 7.046817302703857,
+      "grad_norm": 7.844751834869385,
       "learning_rate": 4.996265870052278e-05,
-      "loss": 1.1917,
+      "loss": 1.2185,
       "step": 300
     },
     {
       "epoch": 0.1,
-      "grad_norm": 8.558070182800293,
+      "grad_norm": 9.986318588256836,
       "learning_rate": 4.977595220313667e-05,
-      "loss": 1.3288,
+      "loss": 1.2979,
       "step": 310
     },
     {
       "epoch": 0.11,
-      "grad_norm": 7.99010705947876,
+      "grad_norm": 7.930599689483643,
       "learning_rate": 4.958924570575056e-05,
-      "loss": 1.1893,
+      "loss": 1.1913,
       "step": 320
     },
     {
       "epoch": 0.11,
-      "grad_norm": 9.63879108428955,
+      "grad_norm": 14.08823013305664,
       "learning_rate": 4.9402539208364454e-05,
-      "loss": 1.1212,
+      "loss": 1.0968,
       "step": 330
     },
     {
       "epoch": 0.11,
-      "grad_norm": 11.649455070495605,
+      "grad_norm": 9.71191120147705,
       "learning_rate": 4.9215832710978346e-05,
-      "loss": 1.0572,
+      "loss": 1.0469,
       "step": 340
     },
     {
       "epoch": 0.12,
-      "grad_norm": 13.81701946258545,
+      "grad_norm": 11.69392204284668,
       "learning_rate": 4.902912621359224e-05,
-      "loss": 1.1869,
+      "loss": 1.152,
       "step": 350
     },
     {
       "epoch": 0.12,
-      "grad_norm": 13.85696029663086,
+      "grad_norm": 12.762391090393066,
       "learning_rate": 4.884241971620613e-05,
-      "loss": 1.5038,
+      "loss": 1.3083,
       "step": 360
     },
     {
       "epoch": 0.12,
-      "grad_norm": 8.543469429016113,
+      "grad_norm": 9.314635276794434,
       "learning_rate": 4.8655713218820016e-05,
-      "loss": 1.0457,
+      "loss": 1.0477,
       "step": 370
     },
     {
       "epoch": 0.13,
-      "grad_norm": 10.270995140075684,
+      "grad_norm": 10.564154624938965,
       "learning_rate": 4.846900672143391e-05,
-      "loss": 0.9543,
+      "loss": 1.0316,
       "step": 380
     },
     {
       "epoch": 0.13,
-      "grad_norm": 8.127942085266113,
+      "grad_norm": 14.425047874450684,
       "learning_rate": 4.82823002240478e-05,
-      "loss": 0.9389,
+      "loss": 1.1457,
       "step": 390
     },
     {
       "epoch": 0.13,
-      "grad_norm": 10.777777671813965,
+      "grad_norm": 11.951525688171387,
       "learning_rate": 4.809559372666169e-05,
-      "loss": 1.0655,
+      "loss": 1.1723,
       "step": 400
     },
     {
       "epoch": 0.14,
-      "grad_norm": 8.693421363830566,
+      "grad_norm": 8.284050941467285,
       "learning_rate": 4.790888722927558e-05,
-      "loss": 1.0275,
+      "loss": 1.0312,
       "step": 410
     },
     {
       "epoch": 0.14,
-      "grad_norm": 10.789031028747559,
+      "grad_norm": 9.24317455291748,
       "learning_rate": 4.772218073188947e-05,
-      "loss": 1.2218,
+      "loss": 1.2024,
       "step": 420
     },
     {
       "epoch": 0.14,
-      "grad_norm": 6.10308313369751,
+      "grad_norm": 5.5853590965271,
       "learning_rate": 4.753547423450336e-05,
-      "loss": 0.9795,
+      "loss": 0.9976,
       "step": 430
     },
     {
       "epoch": 0.15,
-      "grad_norm": 11.924713134765625,
+      "grad_norm": 8.923686027526855,
       "learning_rate": 4.7348767737117256e-05,
-      "loss": 1.0568,
+      "loss": 1.0393,
       "step": 440
     },
     {
       "epoch": 0.15,
-      "grad_norm": 5.780080795288086,
+      "grad_norm": 5.833179950714111,
       "learning_rate": 4.716206123973114e-05,
-      "loss": 0.9916,
+      "loss": 0.8951,
       "step": 450
     },
     {
       "epoch": 0.15,
-      "grad_norm": 7.771975994110107,
+      "grad_norm": 8.868228912353516,
       "learning_rate": 4.697535474234503e-05,
-      "loss": 0.9661,
+      "loss": 0.997,
       "step": 460
     },
     {
       "epoch": 0.16,
-      "grad_norm": 9.780990600585938,
+      "grad_norm": 9.04990005493164,
       "learning_rate": 4.6788648244958926e-05,
-      "loss": 0.884,
+      "loss": 0.8222,
       "step": 470
     },
     {
       "epoch": 0.16,
-      "grad_norm": 6.3119354248046875,
+      "grad_norm": 7.88612699508667,
       "learning_rate": 4.660194174757282e-05,
-      "loss": 1.3266,
+      "loss": 1.2196,
       "step": 480
     },
     {
       "epoch": 0.16,
-      "grad_norm": 6.907146453857422,
+      "grad_norm": 9.290902137756348,
       "learning_rate": 4.6415235250186703e-05,
-      "loss": 1.0728,
+      "loss": 1.1256,
       "step": 490
     },
     {
       "epoch": 0.17,
-      "grad_norm": 6.732804775238037,
+      "grad_norm": 6.916901111602783,
       "learning_rate": 4.6228528752800596e-05,
-      "loss": 1.1793,
+      "loss": 1.2263,
       "step": 500
     },
     {
       "epoch": 0.17,
-      "grad_norm": 5.275418281555176,
+      "grad_norm": 4.902709484100342,
       "learning_rate": 4.604182225541449e-05,
-      "loss": 1.0455,
+      "loss": 0.9711,
       "step": 510
     },
     {
       "epoch": 0.17,
-      "grad_norm": 8.184633255004883,
+      "grad_norm": 9.100408554077148,
       "learning_rate": 4.585511575802838e-05,
-      "loss": 1.2988,
+      "loss": 1.3256,
       "step": 520
     },
     {
       "epoch": 0.18,
-      "grad_norm": 9.226140022277832,
+      "grad_norm": 8.152189254760742,
       "learning_rate": 4.566840926064227e-05,
-      "loss": 0.9656,
+      "loss": 0.9898,
       "step": 530
     },
     {
       "epoch": 0.18,
-      "grad_norm": 9.135764122009277,
+      "grad_norm": 9.004831314086914,
       "learning_rate": 4.5481702763256165e-05,
-      "loss": 0.9574,
+      "loss": 0.9567,
       "step": 540
     },
     {
       "epoch": 0.18,
-      "grad_norm": 9.493541717529297,
+      "grad_norm": 10.926904678344727,
       "learning_rate": 4.529499626587006e-05,
-      "loss": 1.177,
+      "loss": 1.1919,
       "step": 550
     },
     {
       "epoch": 0.19,
-      "grad_norm": 10.075041770935059,
+      "grad_norm": 11.16217041015625,
       "learning_rate": 4.510828976848395e-05,
-      "loss": 1.0754,
+      "loss": 0.9503,
       "step": 560
     },
     {
       "epoch": 0.19,
-      "grad_norm": 9.099665641784668,
+      "grad_norm": 8.947577476501465,
       "learning_rate": 4.492158327109784e-05,
-      "loss": 1.179,
+      "loss": 1.1709,
       "step": 570
     },
     {
       "epoch": 0.19,
-      "grad_norm": 18.290414810180664,
+      "grad_norm": 17.401987075805664,
       "learning_rate": 4.473487677371173e-05,
-      "loss": 1.14,
+      "loss": 1.1109,
       "step": 580
     },
     {
       "epoch": 0.2,
-      "grad_norm": 6.768624305725098,
+      "grad_norm": 4.542303085327148,
       "learning_rate": 4.454817027632562e-05,
-      "loss": 1.0689,
+      "loss": 1.0982,
       "step": 590
     },
     {
       "epoch": 0.2,
-      "grad_norm": 11.730013847351074,
+      "grad_norm": 11.068498611450195,
       "learning_rate": 4.436146377893951e-05,
-      "loss": 0.8808,
+      "loss": 0.8818,
       "step": 600
     },
     {
       "epoch": 0.2,
-      "grad_norm": 7.4615159034729,
+      "grad_norm": 8.869817733764648,
       "learning_rate": 4.4174757281553404e-05,
-      "loss": 1.093,
+      "loss": 0.9708,
       "step": 610
     },
     {
       "epoch": 0.21,
-      "grad_norm": 11.849823951721191,
+      "grad_norm": 11.837583541870117,
       "learning_rate": 4.398805078416729e-05,
-      "loss": 0.9596,
+      "loss": 0.912,
       "step": 620
     },
     {
       "epoch": 0.21,
-      "grad_norm": 8.06859302520752,
+      "grad_norm": 10.15700626373291,
       "learning_rate": 4.380134428678118e-05,
-      "loss": 0.9528,
+      "loss": 1.0116,
       "step": 630
     },
     {
       "epoch": 0.22,
-      "grad_norm": 11.52597713470459,
+      "grad_norm": 14.383340835571289,
       "learning_rate": 4.3614637789395075e-05,
-      "loss": 0.8866,
+      "loss": 0.9178,
       "step": 640
     },
     {
       "epoch": 0.22,
-      "grad_norm": 12.404598236083984,
+      "grad_norm": 10.986087799072266,
       "learning_rate": 4.342793129200897e-05,
-      "loss": 0.7092,
+      "loss": 0.7005,
       "step": 650
     },
     {
       "epoch": 0.22,
-      "grad_norm": 6.062406539916992,
+      "grad_norm": 5.080931186676025,
       "learning_rate": 4.324122479462285e-05,
-      "loss": 0.7407,
+      "loss": 0.834,
       "step": 660
     },
     {
       "epoch": 0.23,
-      "grad_norm": 11.590093612670898,
+      "grad_norm": 11.886563301086426,
       "learning_rate": 4.3054518297236745e-05,
-      "loss": 1.1617,
+      "loss": 1.1476,
       "step": 670
     },
     {
       "epoch": 0.23,
-      "grad_norm": 8.003750801086426,
+      "grad_norm": 6.172848224639893,
       "learning_rate": 4.286781179985064e-05,
-      "loss": 0.9215,
+      "loss": 0.8403,
       "step": 680
     },
     {
       "epoch": 0.23,
-      "grad_norm": 6.014645099639893,
+      "grad_norm": 5.088493824005127,
       "learning_rate": 4.268110530246453e-05,
-      "loss": 0.8719,
+      "loss": 0.9338,
       "step": 690
     },
     {
       "epoch": 0.24,
-      "grad_norm": 7.771167755126953,
+      "grad_norm": 9.035572052001953,
       "learning_rate": 4.2494398805078415e-05,
-      "loss": 0.865,
+      "loss": 0.9069,
       "step": 700
     },
     {
       "epoch": 0.24,
-      "grad_norm": 6.302040100097656,
+      "grad_norm": 6.618842124938965,
       "learning_rate": 4.230769230769231e-05,
-      "loss": 0.7946,
+      "loss": 0.8403,
       "step": 710
     },
     {
       "epoch": 0.24,
-      "grad_norm": 14.395988464355469,
+      "grad_norm": 12.929986953735352,
       "learning_rate": 4.21209858103062e-05,
-      "loss": 1.0984,
+      "loss": 1.0611,
       "step": 720
     },
     {
       "epoch": 0.25,
-      "grad_norm": 7.767952919006348,
+      "grad_norm": 6.48470401763916,
       "learning_rate": 4.193427931292009e-05,
-      "loss": 0.9101,
+      "loss": 0.9453,
       "step": 730
     },
     {
       "epoch": 0.25,
-      "grad_norm": 11.00684642791748,
+      "grad_norm": 13.81301212310791,
       "learning_rate": 4.1747572815533984e-05,
-      "loss": 0.9023,
+      "loss": 0.9737,
       "step": 740
     },
     {
       "epoch": 0.25,
-      "eval_accuracy": 0.6013422818791946,
-      "eval_loss": 1.0584642887115479,
-      "eval_runtime": 1869.6092,
-      "eval_samples_per_second": 0.398,
-      "eval_steps_per_second": 0.05,
+      "eval_accuracy": 0.614765100671141,
+      "eval_loss": 1.0215864181518555,
+      "eval_runtime": 1824.1699,
+      "eval_samples_per_second": 0.408,
+      "eval_steps_per_second": 0.052,
       "step": 745
     },
     {
       "epoch": 1.0,
-      "grad_norm": 8.297401428222656,
+      "grad_norm": 6.812247276306152,
       "learning_rate": 4.1560866318147876e-05,
-      "loss": 0.6975,
+      "loss": 0.7102,
       "step": 750
     },
     {
       "epoch": 1.01,
-      "grad_norm": 8.592920303344727,
+      "grad_norm": 8.490981101989746,
       "learning_rate": 4.137415982076177e-05,
-      "loss": 0.6746,
+      "loss": 0.7269,
       "step": 760
     },
     {
       "epoch": 1.01,
-      "grad_norm": 12.73000717163086,
+      "grad_norm": 14.003782272338867,
       "learning_rate": 4.118745332337566e-05,
-      "loss": 0.7792,
+      "loss": 0.7074,
       "step": 770
     },
     {
       "epoch": 1.01,
-      "grad_norm": 12.524226188659668,
+      "grad_norm": 12.912096977233887,
       "learning_rate": 4.1000746825989546e-05,
-      "loss": 1.1009,
+      "loss": 1.0037,
       "step": 780
     },
     {
       "epoch": 1.02,
-      "grad_norm": 14.503805160522461,
+      "grad_norm": 11.331485748291016,
       "learning_rate": 4.081404032860344e-05,
-      "loss": 0.8554,
+      "loss": 0.8032,
       "step": 790
     },
     {
       "epoch": 1.02,
-      "grad_norm": 6.50175666809082,
+      "grad_norm": 7.185470104217529,
       "learning_rate": 4.062733383121733e-05,
-      "loss": 0.9718,
+      "loss": 0.9243,
       "step": 800
     },
     {
       "epoch": 1.02,
-      "grad_norm": 6.452315807342529,
+      "grad_norm": 6.5612711906433105,
       "learning_rate": 4.0440627333831223e-05,
-      "loss": 0.9421,
+      "loss": 0.874,
       "step": 810
     },
     {
       "epoch": 1.03,
-      "grad_norm": 17.507164001464844,
+      "grad_norm": 14.173837661743164,
       "learning_rate": 4.025392083644511e-05,
-      "loss": 0.9593,
+      "loss": 0.8649,
       "step": 820
     },
     {
       "epoch": 1.03,
-      "grad_norm": 6.719306945800781,
+      "grad_norm": 8.195253372192383,
       "learning_rate": 4.0067214339059e-05,
-      "loss": 0.8568,
+      "loss": 0.7709,
       "step": 830
     },
     {
       "epoch": 1.03,
-      "grad_norm": 8.009960174560547,
+      "grad_norm": 5.4088544845581055,
       "learning_rate": 3.9880507841672894e-05,
-      "loss": 0.796,
+      "loss": 0.7699,
       "step": 840
     },
     {
       "epoch": 1.04,
-      "grad_norm": 6.025060653686523,
+      "grad_norm": 6.817777633666992,
       "learning_rate": 3.9693801344286786e-05,
-      "loss": 0.8486,
+      "loss": 0.8736,
       "step": 850
     },
     {
       "epoch": 1.04,
-      "grad_norm": 8.915096282958984,
+      "grad_norm": 8.318948745727539,
       "learning_rate": 3.950709484690067e-05,
-      "loss": 0.8042,
+      "loss": 0.6866,
       "step": 860
     },
     {
       "epoch": 1.04,
-      "grad_norm": 12.27297592163086,
+      "grad_norm": 12.06480598449707,
       "learning_rate": 3.9320388349514564e-05,
-      "loss": 0.8384,
+      "loss": 0.8834,
       "step": 870
     },
     {
       "epoch": 1.05,
-      "grad_norm": 5.073918342590332,
+      "grad_norm": 6.914089679718018,
       "learning_rate": 3.9133681852128456e-05,
-      "loss": 0.8608,
+      "loss": 0.9787,
       "step": 880
     },
     {
       "epoch": 1.05,
-      "grad_norm": 10.088605880737305,
+      "grad_norm": 7.699115753173828,
       "learning_rate": 3.894697535474235e-05,
-      "loss": 0.8858,
+      "loss": 0.8338,
       "step": 890
     },
     {
       "epoch": 1.05,
-      "grad_norm": 5.86115837097168,
+      "grad_norm": 4.82650089263916,
       "learning_rate": 3.8760268857356234e-05,
-      "loss": 0.6226,
+      "loss": 0.6263,
       "step": 900
     },
     {
       "epoch": 1.06,
-      "grad_norm": 11.275440216064453,
+      "grad_norm": 8.42751407623291,
       "learning_rate": 3.8573562359970126e-05,
-      "loss": 0.8814,
+      "loss": 0.8408,
       "step": 910
     },
     {
       "epoch": 1.06,
-      "grad_norm": 5.241692543029785,
+      "grad_norm": 5.994837760925293,
       "learning_rate": 3.838685586258402e-05,
-      "loss": 0.8163,
+      "loss": 0.8424,
       "step": 920
     },
     {
       "epoch": 1.06,
-      "grad_norm": 8.397137641906738,
+      "grad_norm": 7.990606307983398,
       "learning_rate": 3.820014936519791e-05,
-      "loss": 0.7511,
+      "loss": 0.7187,
       "step": 930
     },
     {
       "epoch": 1.07,
-      "grad_norm": 9.3834810256958,
+      "grad_norm": 10.63245964050293,
       "learning_rate": 3.8013442867811796e-05,
-      "loss": 0.7023,
+      "loss": 0.6456,
       "step": 940
     },
     {
       "epoch": 1.07,
-      "grad_norm": 11.98423957824707,
+      "grad_norm": 12.09002685546875,
       "learning_rate": 3.782673637042569e-05,
-      "loss": 0.7885,
+      "loss": 0.8223,
       "step": 950
     },
     {
       "epoch": 1.07,
-      "grad_norm": 11.273078918457031,
+      "grad_norm": 9.54681396484375,
       "learning_rate": 3.764002987303958e-05,
-      "loss": 0.6545,
+      "loss": 0.7054,
       "step": 960
     },
     {
       "epoch": 1.08,
-      "grad_norm": 14.190473556518555,
+      "grad_norm": 13.245128631591797,
       "learning_rate": 3.745332337565347e-05,
-      "loss": 1.0184,
+      "loss": 0.9493,
       "step": 970
     },
     {
       "epoch": 1.08,
-      "grad_norm": 4.971963405609131,
+      "grad_norm": 4.244826793670654,
       "learning_rate": 3.7266616878267365e-05,
-      "loss": 0.7963,
+      "loss": 0.8649,
       "step": 980
     },
     {
       "epoch": 1.08,
-      "grad_norm": 17.721282958984375,
+      "grad_norm": 10.754876136779785,
       "learning_rate": 3.707991038088126e-05,
-      "loss": 0.9323,
+      "loss": 0.8481,
       "step": 990
     },
     {
       "epoch": 1.09,
-      "grad_norm": 10.512914657592773,
+      "grad_norm": 11.53078842163086,
       "learning_rate": 3.689320388349515e-05,
-      "loss": 0.9391,
+      "loss": 0.9808,
       "step": 1000
     },
     {
       "epoch": 1.09,
-      "grad_norm": 11.076910018920898,
+      "grad_norm": 12.149744033813477,
       "learning_rate": 3.670649738610904e-05,
-      "loss": 0.8941,
+      "loss": 0.8797,
       "step": 1010
     },
     {
       "epoch": 1.09,
-      "grad_norm": 5.027378082275391,
+      "grad_norm": 7.940331935882568,
       "learning_rate": 3.651979088872293e-05,
-      "loss": 0.8346,
+      "loss": 0.8612,
       "step": 1020
     },
     {
       "epoch": 1.1,
-      "grad_norm": 12.940291404724121,
+      "grad_norm": 10.588911056518555,
       "learning_rate": 3.633308439133682e-05,
-      "loss": 0.7896,
+      "loss": 0.7769,
       "step": 1030
     },
     {
       "epoch": 1.1,
-      "grad_norm": 13.178154945373535,
+      "grad_norm": 16.114530563354492,
       "learning_rate": 3.614637789395071e-05,
-      "loss": 0.6817,
+      "loss": 0.7163,
       "step": 1040
     },
     {
       "epoch": 1.1,
-      "grad_norm": 13.505011558532715,
+      "grad_norm": 11.30330753326416,
       "learning_rate": 3.5959671396564605e-05,
-      "loss": 0.7627,
+      "loss": 0.7857,
       "step": 1050
     },
     {
       "epoch": 1.11,
-      "grad_norm": 4.09556245803833,
+      "grad_norm": 4.0768561363220215,
       "learning_rate": 3.577296489917849e-05,
-      "loss": 0.8155,
+      "loss": 0.6938,
       "step": 1060
     },
     {
       "epoch": 1.11,
-      "grad_norm": 12.600196838378906,
+      "grad_norm": 7.4695143699646,
       "learning_rate": 3.558625840179238e-05,
-      "loss": 0.89,
+      "loss": 0.8988,
       "step": 1070
     },
     {
       "epoch": 1.11,
-      "grad_norm": 8.050426483154297,
+      "grad_norm": 11.519210815429688,
       "learning_rate": 3.5399551904406275e-05,
-      "loss": 0.8544,
+      "loss": 0.8402,
       "step": 1080
     },
     {
       "epoch": 1.12,
-      "grad_norm": 13.491654396057129,
+      "grad_norm": 15.348576545715332,
       "learning_rate": 3.521284540702017e-05,
-      "loss": 0.6694,
+      "loss": 0.7148,
       "step": 1090
     },
     {
       "epoch": 1.12,
-      "grad_norm": 6.23304557800293,
+      "grad_norm": 8.274968147277832,
       "learning_rate": 3.502613890963405e-05,
-      "loss": 0.8363,
+      "loss": 0.7931,
       "step": 1100
     },
     {
       "epoch": 1.12,
-      "grad_norm": 12.755766868591309,
+      "grad_norm": 18.91743278503418,
       "learning_rate": 3.4839432412247945e-05,
-      "loss": 0.6294,
+      "loss": 0.6662,
       "step": 1110
     },
     {
       "epoch": 1.13,
-      "grad_norm": 7.085630416870117,
+      "grad_norm": 7.221806049346924,
       "learning_rate": 3.465272591486184e-05,
-      "loss": 0.6517,
+      "loss": 0.6487,
       "step": 1120
     },
     {
       "epoch": 1.13,
-      "grad_norm": 7.111999988555908,
+      "grad_norm": 6.4034013748168945,
       "learning_rate": 3.446601941747573e-05,
-      "loss": 0.6923,
+      "loss": 0.6152,
       "step": 1130
     },
     {
       "epoch": 1.13,
-      "grad_norm": 9.9430513381958,
+      "grad_norm": 8.308573722839355,
       "learning_rate": 3.427931292008962e-05,
-      "loss": 0.7989,
+      "loss": 0.8798,
       "step": 1140
     },
     {
       "epoch": 1.14,
-      "grad_norm": 7.441895484924316,
+      "grad_norm": 7.0101213455200195,
       "learning_rate": 3.409260642270351e-05,
-      "loss": 0.8628,
+      "loss": 0.8473,
       "step": 1150
     },
     {
       "epoch": 1.14,
-      "grad_norm": 11.109057426452637,
+      "grad_norm": 9.986639976501465,
       "learning_rate": 3.39058999253174e-05,
-      "loss": 0.9556,
+      "loss": 0.9895,
       "step": 1160
     },
     {
       "epoch": 1.14,
-      "grad_norm": 3.8683066368103027,
+      "grad_norm": 3.4603047370910645,
       "learning_rate": 3.371919342793129e-05,
-      "loss": 0.7598,
+      "loss": 0.7945,
       "step": 1170
     },
     {
       "epoch": 1.15,
-      "grad_norm": 9.229301452636719,
+      "grad_norm": 9.439742088317871,
       "learning_rate": 3.3532486930545184e-05,
-      "loss": 0.7694,
+      "loss": 0.8037,
       "step": 1180
     },
     {
       "epoch": 1.15,
-      "grad_norm": 6.90978479385376,
+      "grad_norm": 9.237773895263672,
       "learning_rate": 3.334578043315908e-05,
-      "loss": 0.6816,
+      "loss": 0.7623,
       "step": 1190
     },
     {
       "epoch": 1.15,
-      "grad_norm": 10.782751083374023,
+      "grad_norm": 10.11063289642334,
       "learning_rate": 3.315907393577297e-05,
-      "loss": 0.7015,
+      "loss": 0.6762,
       "step": 1200
     },
     {
       "epoch": 1.16,
-      "grad_norm": 12.352110862731934,
+      "grad_norm": 14.077370643615723,
       "learning_rate": 3.297236743838686e-05,
-      "loss": 0.6476,
+      "loss": 0.6943,
       "step": 1210
     },
     {
       "epoch": 1.16,
-      "grad_norm": 12.289131164550781,
+      "grad_norm": 8.673574447631836,
       "learning_rate": 3.2785660941000754e-05,
-      "loss": 0.6243,
+      "loss": 0.6585,
       "step": 1220
     },
     {
       "epoch": 1.16,
-      "grad_norm": 10.164794921875,
+      "grad_norm": 10.235486030578613,
       "learning_rate": 3.259895444361464e-05,
-      "loss": 0.7136,
+      "loss": 0.6866,
       "step": 1230
     },
     {
       "epoch": 1.17,
-      "grad_norm": 11.630318641662598,
+      "grad_norm": 12.640578269958496,
       "learning_rate": 3.241224794622853e-05,
-      "loss": 0.8095,
+      "loss": 0.7851,
       "step": 1240
     },
     {
       "epoch": 1.17,
-      "grad_norm": 8.73193359375,
+      "grad_norm": 6.836172580718994,
       "learning_rate": 3.2225541448842424e-05,
-      "loss": 0.7918,
+      "loss": 0.8098,
       "step": 1250
     },
     {
       "epoch": 1.17,
-      "grad_norm": 10.5596284866333,
+      "grad_norm": 11.899585723876953,
       "learning_rate": 3.2038834951456316e-05,
-      "loss": 0.9087,
+      "loss": 0.9768,
       "step": 1260
     },
     {
       "epoch": 1.18,
-      "grad_norm": 10.88122844696045,
+      "grad_norm": 9.843255043029785,
       "learning_rate": 3.18521284540702e-05,
-      "loss": 0.5758,
+      "loss": 0.5619,
       "step": 1270
     },
     {
       "epoch": 1.18,
-      "grad_norm": 7.157528877258301,
+      "grad_norm": 7.826408863067627,
       "learning_rate": 3.1665421956684094e-05,
-      "loss": 0.645,
+      "loss": 0.7163,
       "step": 1280
     },
     {
       "epoch": 1.18,
-      "grad_norm": 6.619960784912109,
+      "grad_norm": 5.7413482666015625,
       "learning_rate": 3.1478715459297986e-05,
-      "loss": 0.6412,
+      "loss": 0.5594,
       "step": 1290
     },
     {
       "epoch": 1.19,
-      "grad_norm": 9.903483390808105,
+      "grad_norm": 9.495389938354492,
       "learning_rate": 3.129200896191188e-05,
-      "loss": 0.5866,
+      "loss": 0.6122,
       "step": 1300
     },
     {
       "epoch": 1.19,
-      "grad_norm": 8.340232849121094,
+      "grad_norm": 9.183694839477539,
       "learning_rate": 3.1105302464525764e-05,
-      "loss": 0.6971,
+      "loss": 0.7087,
       "step": 1310
     },
     {
       "epoch": 1.19,
-      "grad_norm": 4.8595170974731445,
+      "grad_norm": 2.8270673751831055,
       "learning_rate": 3.0918595967139656e-05,
-      "loss": 0.6824,
+      "loss": 0.6698,
       "step": 1320
     },
     {
       "epoch": 1.2,
-      "grad_norm": 13.451203346252441,
+      "grad_norm": 12.440056800842285,
       "learning_rate": 3.073188946975355e-05,
-      "loss": 0.5937,
+      "loss": 0.5935,
       "step": 1330
     },
     {
       "epoch": 1.2,
-      "grad_norm": 8.88749885559082,
+      "grad_norm": 10.87549114227295,
       "learning_rate": 3.054518297236744e-05,
-      "loss": 0.7642,
+      "loss": 0.756,
       "step": 1340
     },
     {
       "epoch": 1.2,
-      "grad_norm": 10.660982131958008,
+      "grad_norm": 10.946492195129395,
       "learning_rate": 3.035847647498133e-05,
-      "loss": 0.7327,
+      "loss": 0.8023,
       "step": 1350
     },
     {
       "epoch": 1.21,
-      "grad_norm": 12.592216491699219,
+      "grad_norm": 10.33581829071045,
       "learning_rate": 3.0171769977595222e-05,
-      "loss": 0.6769,
+      "loss": 0.672,
       "step": 1360
     },
     {
       "epoch": 1.21,
-      "grad_norm": 3.5256502628326416,
+      "grad_norm": 3.581252336502075,
       "learning_rate": 2.9985063480209115e-05,
-      "loss": 0.6646,
+      "loss": 0.6321,
       "step": 1370
     },
     {
       "epoch": 1.21,
-      "grad_norm": 6.979560852050781,
+      "grad_norm": 6.161603927612305,
       "learning_rate": 2.9798356982823007e-05,
-      "loss": 0.7394,
+      "loss": 0.7259,
       "step": 1380
     },
     {
       "epoch": 1.22,
-      "grad_norm": 8.0765962600708,
+      "grad_norm": 6.92000675201416,
       "learning_rate": 2.9611650485436892e-05,
-      "loss": 0.597,
+      "loss": 0.4861,
       "step": 1390
     },
     {
       "epoch": 1.22,
-      "grad_norm": 6.306230545043945,
+      "grad_norm": 10.259751319885254,
       "learning_rate": 2.9424943988050785e-05,
-      "loss": 0.557,
+      "loss": 0.5506,
       "step": 1400
     },
     {
       "epoch": 1.22,
-      "grad_norm": 7.691712856292725,
+      "grad_norm": 7.796722888946533,
       "learning_rate": 2.9238237490664677e-05,
-      "loss": 0.778,
+      "loss": 0.7225,
       "step": 1410
     },
     {
       "epoch": 1.23,
-      "grad_norm": 10.76540756225586,
+      "grad_norm": 10.522936820983887,
       "learning_rate": 2.905153099327857e-05,
-      "loss": 0.7148,
+      "loss": 0.6722,
       "step": 1420
     },
     {
       "epoch": 1.23,
-      "grad_norm": 8.003518104553223,
+      "grad_norm": 8.046516418457031,
       "learning_rate": 2.8864824495892455e-05,
-      "loss": 0.799,
+      "loss": 0.7846,
       "step": 1430
     },
     {
       "epoch": 1.23,
-      "grad_norm": 8.381891250610352,
+      "grad_norm": 7.680605411529541,
       "learning_rate": 2.8678117998506347e-05,
-      "loss": 0.7274,
+      "loss": 0.6704,
       "step": 1440
     },
     {
       "epoch": 1.24,
-      "grad_norm": 11.402746200561523,
+      "grad_norm": 14.043888092041016,
       "learning_rate": 2.849141150112024e-05,
-      "loss": 0.767,
+      "loss": 0.6614,
       "step": 1450
     },
     {
       "epoch": 1.24,
-      "grad_norm": 12.372224807739258,
+      "grad_norm": 11.873641967773438,
       "learning_rate": 2.8304705003734132e-05,
-      "loss": 0.4786,
+      "loss": 0.5319,
       "step": 1460
     },
     {
       "epoch": 1.24,
-      "grad_norm": 6.330965518951416,
+      "grad_norm": 18.787263870239258,
       "learning_rate": 2.811799850634802e-05,
-      "loss": 0.6594,
+      "loss": 0.6216,
       "step": 1470
     },
     {
       "epoch": 1.25,
-      "grad_norm": 10.601953506469727,
+      "grad_norm": 9.360286712646484,
       "learning_rate": 2.7931292008961913e-05,
-      "loss": 0.5358,
+      "loss": 0.5028,
       "step": 1480
     },
     {
       "epoch": 1.25,
-      "grad_norm": 55.654109954833984,
+      "grad_norm": 49.45521926879883,
       "learning_rate": 2.7744585511575805e-05,
-      "loss": 0.9428,
+      "loss": 0.9205,
       "step": 1490
     },
     {
       "epoch": 1.25,
       "eval_accuracy": 0.6510067114093959,
-      "eval_loss": 0.897102415561676,
-      "eval_runtime": 1724.9592,
-      "eval_samples_per_second": 0.432,
-      "eval_steps_per_second": 0.054,
+      "eval_loss": 0.923988938331604,
+      "eval_runtime": 1688.0484,
+      "eval_samples_per_second": 0.441,
+      "eval_steps_per_second": 0.056,
       "step": 1490
     },
     {
       "epoch": 2.0,
-      "grad_norm": 14.913216590881348,
+      "grad_norm": 11.202278137207031,
       "learning_rate": 2.7557879014189698e-05,
-      "loss": 0.6672,
+      "loss": 0.5834,
       "step": 1500
     },
     {
       "epoch": 2.01,
-      "grad_norm": 8.555597305297852,
+      "grad_norm": 9.430272102355957,
       "learning_rate": 2.7371172516803583e-05,
-      "loss": 0.517,
+      "loss": 0.5599,
       "step": 1510
     },
     {
       "epoch": 2.01,
-      "grad_norm": 13.102532386779785,
+      "grad_norm": 13.628783226013184,
       "learning_rate": 2.7184466019417475e-05,
-      "loss": 0.4576,
+      "loss": 0.5126,
       "step": 1520
     },
     {
       "epoch": 2.01,
-      "grad_norm": 8.147948265075684,
+      "grad_norm": 6.143797397613525,
       "learning_rate": 2.6997759522031368e-05,
-      "loss": 0.5292,
+      "loss": 0.5101,
       "step": 1530
     },
     {
       "epoch": 2.02,
-      "grad_norm": 9.257566452026367,
+      "grad_norm": 9.375954627990723,
       "learning_rate": 2.681105302464526e-05,
-      "loss": 0.5036,
+      "loss": 0.488,
       "step": 1540
     },
     {
       "epoch": 2.02,
-      "grad_norm": 11.560401916503906,
+      "grad_norm": 12.77665901184082,
       "learning_rate": 2.662434652725915e-05,
-      "loss": 0.6044,
+      "loss": 0.6865,
       "step": 1550
     },
     {
       "epoch": 2.02,
-      "grad_norm": 11.875014305114746,
+      "grad_norm": 11.217716217041016,
       "learning_rate": 2.643764002987304e-05,
-      "loss": 0.5306,
+      "loss": 0.5748,
       "step": 1560
     },
     {
       "epoch": 2.03,
-      "grad_norm": 13.541814804077148,
+      "grad_norm": 13.564386367797852,
       "learning_rate": 2.6250933532486934e-05,
-      "loss": 0.6254,
+      "loss": 0.6002,
       "step": 1570
     },
     {
       "epoch": 2.03,
-      "grad_norm": 9.984253883361816,
+      "grad_norm": 9.264657020568848,
       "learning_rate": 2.6064227035100826e-05,
-      "loss": 0.5974,
+      "loss": 0.615,
       "step": 1580
     },
     {
       "epoch": 2.03,
-      "grad_norm": 10.722140312194824,
+      "grad_norm": 15.021964073181152,
       "learning_rate": 2.587752053771471e-05,
-      "loss": 0.7041,
+      "loss": 0.7611,
       "step": 1590
     },
     {
       "epoch": 2.04,
-      "grad_norm": 10.557269096374512,
+      "grad_norm": 8.8477783203125,
       "learning_rate": 2.5690814040328604e-05,
-      "loss": 0.5909,
+      "loss": 0.5559,
       "step": 1600
     },
     {
       "epoch": 2.04,
-      "grad_norm": 11.491299629211426,
+      "grad_norm": 12.896989822387695,
       "learning_rate": 2.5504107542942496e-05,
-      "loss": 0.4946,
+      "loss": 0.5796,
       "step": 1610
     },
     {
       "epoch": 2.04,
-      "grad_norm": 8.835367202758789,
+      "grad_norm": 5.042093276977539,
       "learning_rate": 2.531740104555639e-05,
-      "loss": 0.6644,
+      "loss": 0.632,
       "step": 1620
     },
     {
       "epoch": 2.05,
-      "grad_norm": 11.745617866516113,
+      "grad_norm": 9.28765869140625,
       "learning_rate": 2.5130694548170274e-05,
-      "loss": 0.6128,
+      "loss": 0.6429,
       "step": 1630
     },
     {
       "epoch": 2.05,
-      "grad_norm": 13.023053169250488,
+      "grad_norm": 13.730548858642578,
       "learning_rate": 2.4943988050784166e-05,
-      "loss": 0.6555,
+      "loss": 0.6212,
       "step": 1640
     },
     {
       "epoch": 2.05,
-      "grad_norm": 6.082352161407471,
+      "grad_norm": 5.59623908996582,
       "learning_rate": 2.475728155339806e-05,
-      "loss": 0.5754,
+      "loss": 0.5832,
       "step": 1650
     },
     {
       "epoch": 2.06,
-      "grad_norm": 4.05959415435791,
+      "grad_norm": 6.121994972229004,
       "learning_rate": 2.4570575056011947e-05,
-      "loss": 0.4325,
+      "loss": 0.5236,
       "step": 1660
     },
     {
       "epoch": 2.06,
-      "grad_norm": 14.252350807189941,
+      "grad_norm": 14.531633377075195,
       "learning_rate": 2.4383868558625843e-05,
-      "loss": 0.4956,
+      "loss": 0.5765,
       "step": 1670
     },
     {
       "epoch": 2.06,
-      "grad_norm": 6.615436553955078,
+      "grad_norm": 6.720921993255615,
       "learning_rate": 2.4197162061239732e-05,
-      "loss": 0.5474,
+      "loss": 0.6104,
       "step": 1680
     },
     {
       "epoch": 2.07,
-      "grad_norm": 7.025275230407715,
+      "grad_norm": 6.224832057952881,
       "learning_rate": 2.4010455563853624e-05,
-      "loss": 0.6072,
+      "loss": 0.6797,
       "step": 1690
     },
     {
       "epoch": 2.07,
-      "grad_norm": 15.101826667785645,
+      "grad_norm": 10.963885307312012,
       "learning_rate": 2.3823749066467517e-05,
-      "loss": 0.6705,
+      "loss": 0.579,
       "step": 1700
     },
     {
       "epoch": 2.07,
-      "grad_norm": 11.996399879455566,
+      "grad_norm": 9.707019805908203,
       "learning_rate": 2.3637042569081406e-05,
-      "loss": 0.5326,
+      "loss": 0.5891,
       "step": 1710
     },
     {
       "epoch": 2.08,
-      "grad_norm": 11.746566772460938,
+      "grad_norm": 12.264431953430176,
       "learning_rate": 2.3450336071695298e-05,
-      "loss": 0.6512,
+      "loss": 0.6558,
       "step": 1720
     },
     {
       "epoch": 2.08,
-      "grad_norm": 7.299741744995117,
+      "grad_norm": 9.180527687072754,
       "learning_rate": 2.3263629574309187e-05,
-      "loss": 0.4751,
+      "loss": 0.5394,
       "step": 1730
     },
     {
       "epoch": 2.08,
-      "grad_norm": 10.497876167297363,
+      "grad_norm": 10.08325481414795,
       "learning_rate": 2.307692307692308e-05,
-      "loss": 0.7083,
+      "loss": 0.6894,
       "step": 1740
     },
     {
       "epoch": 2.09,
-      "grad_norm": 6.633430480957031,
+      "grad_norm": 7.65585994720459,
       "learning_rate": 2.2890216579536968e-05,
-      "loss": 0.4854,
+      "loss": 0.4734,
       "step": 1750
     },
     {
       "epoch": 2.09,
-      "grad_norm": 8.004325866699219,
+      "grad_norm": 6.85057258605957,
       "learning_rate": 2.270351008215086e-05,
-      "loss": 0.5795,
+      "loss": 0.5201,
       "step": 1760
     },
     {
       "epoch": 2.09,
-      "grad_norm": 9.713132858276367,
+      "grad_norm": 9.979317665100098,
       "learning_rate": 2.251680358476475e-05,
-      "loss": 0.5735,
+      "loss": 0.5334,
       "step": 1770
     },
     {
       "epoch": 2.1,
-      "grad_norm": 6.221426963806152,
+      "grad_norm": 7.880558013916016,
       "learning_rate": 2.233009708737864e-05,
-      "loss": 0.5405,
+      "loss": 0.6427,
       "step": 1780
     },
     {
       "epoch": 2.1,
-      "grad_norm": 6.872443675994873,
+      "grad_norm": 7.206289291381836,
       "learning_rate": 2.2143390589992534e-05,
-      "loss": 0.7661,
+      "loss": 0.7938,
       "step": 1790
     },
     {
       "epoch": 2.1,
-      "grad_norm": 6.894876956939697,
+      "grad_norm": 4.514979839324951,
       "learning_rate": 2.1956684092606426e-05,
-      "loss": 0.4521,
+      "loss": 0.4731,
       "step": 1800
     },
     {
       "epoch": 2.11,
-      "grad_norm": 3.6292059421539307,
+      "grad_norm": 5.672696590423584,
       "learning_rate": 2.1769977595220315e-05,
-      "loss": 0.5084,
+      "loss": 0.5509,
       "step": 1810
     },
     {
       "epoch": 2.11,
-      "grad_norm": 13.099087715148926,
+      "grad_norm": 12.793485641479492,
       "learning_rate": 2.1583271097834207e-05,
-      "loss": 0.644,
+      "loss": 0.6027,
       "step": 1820
     },
     {
       "epoch": 2.11,
-      "grad_norm": 13.508889198303223,
+      "grad_norm": 12.888551712036133,
       "learning_rate": 2.1396564600448096e-05,
-      "loss": 0.6271,
+      "loss": 0.6495,
       "step": 1830
     },
     {
       "epoch": 2.12,
-      "grad_norm": 9.162331581115723,
+      "grad_norm": 8.044211387634277,
       "learning_rate": 2.120985810306199e-05,
-      "loss": 0.5586,
+      "loss": 0.5304,
       "step": 1840
     },
     {
       "epoch": 2.12,
-      "grad_norm": 9.340535163879395,
+      "grad_norm": 11.73928451538086,
       "learning_rate": 2.1023151605675877e-05,
-      "loss": 0.6222,
+      "loss": 0.6466,
       "step": 1850
     },
     {
       "epoch": 2.12,
-      "grad_norm": 10.59788990020752,
+      "grad_norm": 9.793115615844727,
       "learning_rate": 2.083644510828977e-05,
-      "loss": 0.6439,
+      "loss": 0.6457,
       "step": 1860
     },
     {
       "epoch": 2.13,
-      "grad_norm": 13.132092475891113,
+      "grad_norm": 15.4037504196167,
       "learning_rate": 2.064973861090366e-05,
-      "loss": 0.5401,
+      "loss": 0.6134,
       "step": 1870
     },
     {
       "epoch": 2.13,
-      "grad_norm": 11.530851364135742,
+      "grad_norm": 8.58609390258789,
       "learning_rate": 2.046303211351755e-05,
-      "loss": 0.6462,
+      "loss": 0.6056,
       "step": 1880
     },
     {
       "epoch": 2.13,
-      "grad_norm": 15.348837852478027,
+      "grad_norm": 14.316490173339844,
       "learning_rate": 2.0276325616131443e-05,
-      "loss": 0.6606,
+      "loss": 0.6665,
       "step": 1890
     },
     {
       "epoch": 2.14,
-      "grad_norm": 16.934356689453125,
+      "grad_norm": 15.085420608520508,
       "learning_rate": 2.0089619118745336e-05,
-      "loss": 0.6352,
+      "loss": 0.6414,
       "step": 1900
     },
     {
       "epoch": 2.14,
-      "grad_norm": 9.712738037109375,
+      "grad_norm": 10.707281112670898,
       "learning_rate": 1.9902912621359225e-05,
-      "loss": 0.8095,
+      "loss": 0.7794,
       "step": 1910
     },
     {
       "epoch": 2.14,
-      "grad_norm": 10.83343505859375,
+      "grad_norm": 13.632383346557617,
       "learning_rate": 1.9716206123973117e-05,
-      "loss": 0.5243,
+      "loss": 0.534,
       "step": 1920
     },
     {
       "epoch": 2.15,
-      "grad_norm": 10.138250350952148,
+      "grad_norm": 8.533977508544922,
       "learning_rate": 1.9529499626587006e-05,
-      "loss": 0.4842,
+      "loss": 0.549,
       "step": 1930
     },
     {
       "epoch": 2.15,
-      "grad_norm": 9.77036190032959,
+      "grad_norm": 10.36202335357666,
       "learning_rate": 1.9342793129200898e-05,
-      "loss": 0.6791,
+      "loss": 0.6243,
       "step": 1940
     },
     {
       "epoch": 2.15,
-      "grad_norm": 9.89406967163086,
+      "grad_norm": 9.732759475708008,
       "learning_rate": 1.9156086631814787e-05,
-      "loss": 0.5933,
+      "loss": 0.5558,
       "step": 1950
     },
     {
       "epoch": 2.16,
-      "grad_norm": 7.667896747589111,
+      "grad_norm": 11.463001251220703,
       "learning_rate": 1.896938013442868e-05,
-      "loss": 0.558,
+      "loss": 0.608,
       "step": 1960
     },
     {
       "epoch": 2.16,
-      "grad_norm": 21.039913177490234,
+      "grad_norm": 12.787510871887207,
       "learning_rate": 1.8782673637042568e-05,
-      "loss": 0.5028,
+      "loss": 0.4653,
       "step": 1970
     },
     {
       "epoch": 2.16,
-      "grad_norm": 3.437727689743042,
+      "grad_norm": 2.2048239707946777,
       "learning_rate": 1.859596713965646e-05,
-      "loss": 0.4429,
+      "loss": 0.4185,
       "step": 1980
     },
     {
       "epoch": 2.17,
-      "grad_norm": 12.366796493530273,
+      "grad_norm": 14.132601737976074,
       "learning_rate": 1.8409260642270353e-05,
-      "loss": 0.6008,
+      "loss": 0.5964,
       "step": 1990
     },
     {
       "epoch": 2.17,
-      "grad_norm": 17.259078979492188,
+      "grad_norm": 13.3015775680542,
       "learning_rate": 1.8222554144884245e-05,
-      "loss": 0.5279,
+      "loss": 0.5275,
       "step": 2000
     },
     {
       "epoch": 2.17,
-      "grad_norm": 8.58683967590332,
+      "grad_norm": 9.547086715698242,
       "learning_rate": 1.8035847647498134e-05,
-      "loss": 0.4038,
+      "loss": 0.4277,
       "step": 2010
     },
     {
       "epoch": 2.18,
-      "grad_norm": 6.0911970138549805,
+      "grad_norm": 4.934826374053955,
       "learning_rate": 1.7849141150112026e-05,
-      "loss": 0.5495,
+      "loss": 0.5148,
       "step": 2020
     },
     {
       "epoch": 2.18,
-      "grad_norm": 9.944643020629883,
+      "grad_norm": 8.04011344909668,
       "learning_rate": 1.7662434652725915e-05,
-      "loss": 0.5382,
+      "loss": 0.5389,
       "step": 2030
     },
     {
       "epoch": 2.18,
-      "grad_norm": 13.859691619873047,
+      "grad_norm": 6.342478275299072,
       "learning_rate": 1.7475728155339808e-05,
-      "loss": 0.6927,
+      "loss": 0.6957,
       "step": 2040
     },
     {
       "epoch": 2.19,
-      "grad_norm": 7.948896408081055,
+      "grad_norm": 8.864142417907715,
       "learning_rate": 1.7289021657953697e-05,
-      "loss": 0.3874,
+      "loss": 0.4428,
       "step": 2050
     },
     {
       "epoch": 2.19,
-      "grad_norm": 6.207942485809326,
+      "grad_norm": 5.753154277801514,
       "learning_rate": 1.710231516056759e-05,
-      "loss": 0.4023,
+      "loss": 0.3995,
       "step": 2060
     },
     {
       "epoch": 2.19,
-      "grad_norm": 5.350401401519775,
+      "grad_norm": 5.699671745300293,
       "learning_rate": 1.6915608663181478e-05,
-      "loss": 0.5397,
+      "loss": 0.4798,
       "step": 2070
     },
     {
       "epoch": 2.2,
-      "grad_norm": 10.37016487121582,
+      "grad_norm": 10.056828498840332,
       "learning_rate": 1.672890216579537e-05,
-      "loss": 0.5941,
+      "loss": 0.6173,
       "step": 2080
     },
     {
       "epoch": 2.2,
-      "grad_norm": 9.93877124786377,
+      "grad_norm": 6.651643753051758,
       "learning_rate": 1.654219566840926e-05,
-      "loss": 0.4433,
+      "loss": 0.4855,
       "step": 2090
     },
     {
       "epoch": 2.2,
-      "grad_norm": 14.147184371948242,
+      "grad_norm": 10.440901756286621,
       "learning_rate": 1.635548917102315e-05,
-      "loss": 0.6687,
+      "loss": 0.5887,
       "step": 2100
     },
     {
       "epoch": 2.21,
-      "grad_norm": 2.5294337272644043,
+      "grad_norm": 5.655547618865967,
       "learning_rate": 1.6168782673637044e-05,
-      "loss": 0.5144,
+      "loss": 0.5435,
       "step": 2110
     },
     {
       "epoch": 2.21,
-      "grad_norm": 14.119583129882812,
+      "grad_norm": 10.561293601989746,
       "learning_rate": 1.5982076176250936e-05,
-      "loss": 0.4195,
+      "loss": 0.3069,
       "step": 2120
     },
     {
       "epoch": 2.22,
-      "grad_norm": 5.246395587921143,
+      "grad_norm": 10.159906387329102,
       "learning_rate": 1.5795369678864825e-05,
-      "loss": 0.5291,
+      "loss": 0.5445,
       "step": 2130
     },
     {
       "epoch": 2.22,
-      "grad_norm": 5.959507465362549,
+      "grad_norm": 3.5584466457366943,
       "learning_rate": 1.5608663181478717e-05,
-      "loss": 0.4513,
+      "loss": 0.3836,
       "step": 2140
     },
     {
       "epoch": 2.22,
-      "grad_norm": 12.30712604522705,
+      "grad_norm": 15.946599960327148,
       "learning_rate": 1.5421956684092606e-05,
-      "loss": 0.5737,
+      "loss": 0.5615,
       "step": 2150
     },
     {
       "epoch": 2.23,
-      "grad_norm": 13.365155220031738,
+      "grad_norm": 11.349198341369629,
       "learning_rate": 1.5235250186706498e-05,
-      "loss": 0.6143,
+      "loss": 0.6222,
       "step": 2160
     },
     {
       "epoch": 2.23,
-      "grad_norm": 15.408020973205566,
+      "grad_norm": 11.590112686157227,
       "learning_rate": 1.5048543689320387e-05,
-      "loss": 0.4525,
+      "loss": 0.4596,
       "step": 2170
     },
     {
       "epoch": 2.23,
-      "grad_norm": 8.355168342590332,
+      "grad_norm": 6.031866073608398,
       "learning_rate": 1.486183719193428e-05,
-      "loss": 0.6192,
+      "loss": 0.6662,
       "step": 2180
     },
     {
       "epoch": 2.24,
-      "grad_norm": 14.046072006225586,
+      "grad_norm": 11.580230712890625,
       "learning_rate": 1.467513069454817e-05,
-      "loss": 0.5334,
+      "loss": 0.5664,
       "step": 2190
     },
     {
       "epoch": 2.24,
-      "grad_norm": 6.0648908615112305,
+      "grad_norm": 7.525058746337891,
       "learning_rate": 1.4488424197162062e-05,
-      "loss": 0.6275,
+      "loss": 0.5321,
       "step": 2200
     },
     {
       "epoch": 2.24,
-      "grad_norm": 9.870409965515137,
+      "grad_norm": 10.779327392578125,
       "learning_rate": 1.4301717699775951e-05,
-      "loss": 0.5459,
+      "loss": 0.5125,
       "step": 2210
     },
     {
       "epoch": 2.25,
-      "grad_norm": 10.847134590148926,
+      "grad_norm": 10.189163208007812,
       "learning_rate": 1.4115011202389844e-05,
-      "loss": 0.3645,
+      "loss": 0.3914,
       "step": 2220
     },
     {
       "epoch": 2.25,
-      "grad_norm": 7.336729049682617,
+      "grad_norm": 10.56506633758545,
       "learning_rate": 1.3928304705003734e-05,
-      "loss": 0.4575,
+      "loss": 0.456,
       "step": 2230
     },
     {
       "epoch": 2.25,
-      "eval_accuracy": 0.7181208053691275,
-      "eval_loss": 0.7427847981452942,
-      "eval_runtime": 1683.7503,
-      "eval_samples_per_second": 0.442,
-      "eval_steps_per_second": 0.056,
+      "eval_accuracy": 0.7154362416107383,
+      "eval_loss": 0.7563889622688293,
+      "eval_runtime": 1697.7729,
+      "eval_samples_per_second": 0.439,
+      "eval_steps_per_second": 0.055,
       "step": 2235
     },
     {
       "epoch": 3.0,
-      "grad_norm": 8.262899398803711,
+      "grad_norm": 6.929454326629639,
       "learning_rate": 1.3741598207617627e-05,
-      "loss": 0.6098,
+      "loss": 0.6988,
       "step": 2240
     },
     {
       "epoch": 3.01,
-      "grad_norm": 11.888318061828613,
+      "grad_norm": 14.029314994812012,
       "learning_rate": 1.3554891710231516e-05,
-      "loss": 0.6722,
+      "loss": 0.5992,
       "step": 2250
     },
     {
       "epoch": 3.01,
-      "grad_norm": 4.9109320640563965,
+      "grad_norm": 5.4998626708984375,
       "learning_rate": 1.3368185212845408e-05,
-      "loss": 0.302,
+      "loss": 0.2611,
       "step": 2260
     },
     {
       "epoch": 3.01,
-      "grad_norm": 3.660041570663452,
+      "grad_norm": 8.145641326904297,
       "learning_rate": 1.3181478715459297e-05,
-      "loss": 0.4001,
+      "loss": 0.4163,
       "step": 2270
     },
     {
       "epoch": 3.02,
-      "grad_norm": 5.666093349456787,
+      "grad_norm": 9.244057655334473,
       "learning_rate": 1.2994772218073189e-05,
-      "loss": 0.4406,
+      "loss": 0.4883,
       "step": 2280
     },
     {
       "epoch": 3.02,
-      "grad_norm": 5.51093053817749,
+      "grad_norm": 5.202174186706543,
       "learning_rate": 1.2808065720687081e-05,
-      "loss": 0.434,
+      "loss": 0.47,
       "step": 2290
     },
     {
       "epoch": 3.02,
-      "grad_norm": 11.179670333862305,
+      "grad_norm": 9.33508586883545,
       "learning_rate": 1.2621359223300972e-05,
-      "loss": 0.4668,
+      "loss": 0.4289,
       "step": 2300
     },
     {
       "epoch": 3.03,
-      "grad_norm": 8.03001880645752,
+      "grad_norm": 7.724455833435059,
       "learning_rate": 1.2434652725914863e-05,
-      "loss": 0.4569,
+      "loss": 0.4514,
       "step": 2310
     },
     {
       "epoch": 3.03,
-      "grad_norm": 8.949047088623047,
+      "grad_norm": 4.528939723968506,
       "learning_rate": 1.2247946228528753e-05,
-      "loss": 0.4284,
+      "loss": 0.4015,
       "step": 2320
     },
     {
       "epoch": 3.03,
-      "grad_norm": 10.540315628051758,
+      "grad_norm": 7.545526027679443,
       "learning_rate": 1.2061239731142644e-05,
-      "loss": 0.5279,
+      "loss": 0.5546,
       "step": 2330
     },
     {
       "epoch": 3.04,
-      "grad_norm": 14.215402603149414,
+      "grad_norm": 12.792080879211426,
       "learning_rate": 1.1874533233756534e-05,
-      "loss": 0.4004,
+      "loss": 0.4298,
       "step": 2340
     },
     {
       "epoch": 3.04,
-      "grad_norm": 18.047544479370117,
+      "grad_norm": 12.657350540161133,
       "learning_rate": 1.1687826736370427e-05,
-      "loss": 0.4778,
+      "loss": 0.4234,
       "step": 2350
     },
     {
       "epoch": 3.04,
-      "grad_norm": 9.015625953674316,
+      "grad_norm": 10.286149024963379,
       "learning_rate": 1.1501120238984317e-05,
-      "loss": 0.5168,
+      "loss": 0.5744,
       "step": 2360
     },
     {
       "epoch": 3.05,
-      "grad_norm": 11.635021209716797,
+      "grad_norm": 12.716216087341309,
       "learning_rate": 1.1314413741598208e-05,
-      "loss": 0.4268,
+      "loss": 0.447,
       "step": 2370
     },
     {
       "epoch": 3.05,
-      "grad_norm": 7.7300004959106445,
+      "grad_norm": 10.211912155151367,
       "learning_rate": 1.1127707244212099e-05,
-      "loss": 0.4186,
+      "loss": 0.3881,
       "step": 2380
     },
     {
       "epoch": 3.05,
-      "grad_norm": 8.837210655212402,
+      "grad_norm": 9.413333892822266,
       "learning_rate": 1.094100074682599e-05,
-      "loss": 0.555,
+      "loss": 0.5085,
       "step": 2390
     },
     {
       "epoch": 3.06,
-      "grad_norm": 6.158858299255371,
+      "grad_norm": 5.344712734222412,
       "learning_rate": 1.0754294249439881e-05,
-      "loss": 0.5783,
+      "loss": 0.5566,
       "step": 2400
     },
     {
       "epoch": 3.06,
-      "grad_norm": 7.499691009521484,
+      "grad_norm": 7.993694305419922,
       "learning_rate": 1.0567587752053772e-05,
-      "loss": 0.4923,
+      "loss": 0.53,
       "step": 2410
     },
     {
       "epoch": 3.06,
-      "grad_norm": 12.26675796508789,
+      "grad_norm": 11.350872993469238,
       "learning_rate": 1.0380881254667663e-05,
-      "loss": 0.3842,
+      "loss": 0.3918,
       "step": 2420
     },
     {
       "epoch": 3.07,
-      "grad_norm": 11.513700485229492,
+      "grad_norm": 13.153114318847656,
       "learning_rate": 1.0194174757281553e-05,
-      "loss": 0.3955,
+      "loss": 0.3662,
       "step": 2430
     },
     {
       "epoch": 3.07,
-      "grad_norm": 5.970608234405518,
+      "grad_norm": 5.188636302947998,
       "learning_rate": 1.0007468259895444e-05,
-      "loss": 0.3256,
+      "loss": 0.3685,
       "step": 2440
     },
     {
       "epoch": 3.07,
-      "grad_norm": 10.886767387390137,
+      "grad_norm": 10.718949317932129,
       "learning_rate": 9.820761762509336e-06,
-      "loss": 0.3808,
+      "loss": 0.3139,
       "step": 2450
     },
     {
       "epoch": 3.08,
-      "grad_norm": 5.535740375518799,
+      "grad_norm": 5.5543131828308105,
       "learning_rate": 9.634055265123227e-06,
-      "loss": 0.4004,
+      "loss": 0.3762,
       "step": 2460
     },
     {
       "epoch": 3.08,
-      "grad_norm": 7.606982231140137,
+      "grad_norm": 12.107147216796875,
       "learning_rate": 9.447348767737117e-06,
-      "loss": 0.492,
+      "loss": 0.4458,
       "step": 2470
     },
     {
       "epoch": 3.08,
-      "grad_norm": 4.145596981048584,
+      "grad_norm": 8.29355525970459,
       "learning_rate": 9.260642270351008e-06,
-      "loss": 0.3662,
+      "loss": 0.3784,
       "step": 2480
     },
     {
       "epoch": 3.09,
-      "grad_norm": 5.0407490730285645,
+      "grad_norm": 4.374119758605957,
       "learning_rate": 9.073935772964899e-06,
-      "loss": 0.2492,
+      "loss": 0.2344,
       "step": 2490
     },
     {
       "epoch": 3.09,
-      "grad_norm": 7.193727016448975,
+      "grad_norm": 7.9884772300720215,
       "learning_rate": 8.88722927557879e-06,
-      "loss": 0.3642,
+      "loss": 0.2879,
       "step": 2500
     },
     {
       "epoch": 3.09,
-      "grad_norm": 6.277304172515869,
+      "grad_norm": 5.281538009643555,
       "learning_rate": 8.700522778192682e-06,
-      "loss": 0.4605,
+      "loss": 0.4637,
       "step": 2510
     },
     {
       "epoch": 3.1,
-      "grad_norm": 10.555391311645508,
+      "grad_norm": 11.596240997314453,
       "learning_rate": 8.513816280806572e-06,
-      "loss": 0.3938,
+      "loss": 0.3851,
       "step": 2520
     },
     {
       "epoch": 3.1,
-      "grad_norm": 5.211946964263916,
+      "grad_norm": 4.9211859703063965,
       "learning_rate": 8.327109783420463e-06,
-      "loss": 0.3817,
+      "loss": 0.4068,
       "step": 2530
     },
     {
       "epoch": 3.1,
-      "grad_norm": 11.565357208251953,
+      "grad_norm": 17.217477798461914,
       "learning_rate": 8.140403286034353e-06,
-      "loss": 0.3402,
+      "loss": 0.4238,
       "step": 2540
     },
     {
       "epoch": 3.11,
-      "grad_norm": 7.738372802734375,
+      "grad_norm": 5.638582229614258,
       "learning_rate": 7.953696788648244e-06,
-      "loss": 0.555,
+      "loss": 0.5215,
       "step": 2550
     },
     {
       "epoch": 3.11,
-      "grad_norm": 5.373858451843262,
+      "grad_norm": 5.49488639831543,
       "learning_rate": 7.766990291262136e-06,
-      "loss": 0.4478,
+      "loss": 0.4626,
       "step": 2560
     },
     {
       "epoch": 3.11,
-      "grad_norm": 11.583991050720215,
+      "grad_norm": 11.323406219482422,
       "learning_rate": 7.580283793876028e-06,
-      "loss": 0.3699,
+      "loss": 0.385,
       "step": 2570
     },
     {
       "epoch": 3.12,
-      "grad_norm": 7.290119171142578,
+      "grad_norm": 7.352632522583008,
       "learning_rate": 7.393577296489919e-06,
-      "loss": 0.4119,
+      "loss": 0.4436,
       "step": 2580
     },
     {
       "epoch": 3.12,
-      "grad_norm": 13.199901580810547,
+      "grad_norm": 9.734176635742188,
       "learning_rate": 7.20687079910381e-06,
-      "loss": 0.3973,
+      "loss": 0.3708,
       "step": 2590
     },
     {
       "epoch": 3.12,
-      "grad_norm": 5.801746368408203,
+      "grad_norm": 7.2649149894714355,
       "learning_rate": 7.0201643017177005e-06,
-      "loss": 0.278,
+      "loss": 0.3387,
       "step": 2600
     },
     {
       "epoch": 3.13,
-      "grad_norm": 7.40468692779541,
+      "grad_norm": 6.474803447723389,
       "learning_rate": 6.833457804331592e-06,
-      "loss": 0.4457,
+      "loss": 0.4337,
       "step": 2610
     },
     {
       "epoch": 3.13,
-      "grad_norm": 7.056453227996826,
+      "grad_norm": 7.163308143615723,
       "learning_rate": 6.6467513069454825e-06,
-      "loss": 0.3933,
+      "loss": 0.353,
       "step": 2620
     },
     {
       "epoch": 3.13,
-      "grad_norm": 1.5562771558761597,
+      "grad_norm": 1.612182378768921,
       "learning_rate": 6.460044809559373e-06,
-      "loss": 0.3659,
+      "loss": 0.3491,
       "step": 2630
     },
     {
       "epoch": 3.14,
-      "grad_norm": 12.124537467956543,
+      "grad_norm": 10.161432266235352,
       "learning_rate": 6.273338312173265e-06,
-      "loss": 0.417,
+      "loss": 0.3641,
       "step": 2640
     },
     {
       "epoch": 3.14,
-      "grad_norm": 8.535222053527832,
+      "grad_norm": 10.15164566040039,
       "learning_rate": 6.086631814787154e-06,
-      "loss": 0.4438,
+      "loss": 0.5455,
       "step": 2650
     },
     {
       "epoch": 3.14,
-      "grad_norm": 14.003257751464844,
+      "grad_norm": 7.675275802612305,
       "learning_rate": 5.899925317401046e-06,
-      "loss": 0.4157,
+      "loss": 0.3785,
       "step": 2660
     },
     {
       "epoch": 3.15,
-      "grad_norm": 8.428786277770996,
+      "grad_norm": 9.649803161621094,
       "learning_rate": 5.7132188200149364e-06,
-      "loss": 0.3332,
+      "loss": 0.358,
       "step": 2670
     },
     {
       "epoch": 3.15,
-      "grad_norm": 7.349361896514893,
+      "grad_norm": 6.838894844055176,
       "learning_rate": 5.526512322628828e-06,
-      "loss": 0.354,
+      "loss": 0.3493,
       "step": 2680
     },
     {
       "epoch": 3.15,
-      "grad_norm": 9.289456367492676,
+      "grad_norm": 8.323789596557617,
       "learning_rate": 5.3398058252427185e-06,
-      "loss": 0.4407,
+      "loss": 0.3981,
       "step": 2690
     },
     {
       "epoch": 3.16,
-      "grad_norm": 6.27097225189209,
+      "grad_norm": 5.8700761795043945,
       "learning_rate": 5.153099327856609e-06,
-      "loss": 0.2596,
+      "loss": 0.2634,
       "step": 2700
     },
     {
       "epoch": 3.16,
-      "grad_norm": 15.719304084777832,
+      "grad_norm": 13.782732009887695,
       "learning_rate": 4.966392830470501e-06,
-      "loss": 0.5207,
+      "loss": 0.5584,
       "step": 2710
     },
     {
       "epoch": 3.16,
-      "grad_norm": 7.534648895263672,
+      "grad_norm": 10.914461135864258,
       "learning_rate": 4.779686333084392e-06,
-      "loss": 0.3938,
+      "loss": 0.455,
       "step": 2720
     },
     {
       "epoch": 3.17,
-      "grad_norm": 8.981090545654297,
+      "grad_norm": 11.195931434631348,
       "learning_rate": 4.592979835698283e-06,
-      "loss": 0.248,
+      "loss": 0.2783,
       "step": 2730
     },
     {
       "epoch": 3.17,
-      "grad_norm": 11.375850677490234,
+      "grad_norm": 11.526093482971191,
       "learning_rate": 4.406273338312174e-06,
-      "loss": 0.2799,
+      "loss": 0.2719,
       "step": 2740
     },
     {
       "epoch": 3.17,
-      "grad_norm": 12.491207122802734,
+      "grad_norm": 12.902427673339844,
       "learning_rate": 4.219566840926065e-06,
-      "loss": 0.4839,
+      "loss": 0.4308,
       "step": 2750
     },
     {
       "epoch": 3.18,
-      "grad_norm": 5.428218364715576,
+      "grad_norm": 7.28366756439209,
       "learning_rate": 4.032860343539955e-06,
-      "loss": 0.4289,
+      "loss": 0.4471,
       "step": 2760
     },
     {
       "epoch": 3.18,
-      "grad_norm": 7.776615142822266,
+      "grad_norm": 7.770831108093262,
       "learning_rate": 3.846153846153847e-06,
-      "loss": 0.4563,
+      "loss": 0.5488,
       "step": 2770
     },
     {
       "epoch": 3.18,
-      "grad_norm": 4.943689823150635,
+      "grad_norm": 5.66066837310791,
       "learning_rate": 3.6594473487677374e-06,
-      "loss": 0.3819,
+      "loss": 0.4157,
       "step": 2780
     },
     {
       "epoch": 3.19,
-      "grad_norm": 10.917070388793945,
+      "grad_norm": 4.850650787353516,
       "learning_rate": 3.4727408513816284e-06,
-      "loss": 0.4038,
+      "loss": 0.375,
       "step": 2790
     },
     {
       "epoch": 3.19,
-      "grad_norm": 7.394994258880615,
+      "grad_norm": 7.98192024230957,
       "learning_rate": 3.2860343539955195e-06,
-      "loss": 0.4088,
+      "loss": 0.4436,
       "step": 2800
     },
     {
       "epoch": 3.19,
-      "grad_norm": 6.575070381164551,
+      "grad_norm": 9.876733779907227,
       "learning_rate": 3.09932785660941e-06,
-      "loss": 0.6557,
+      "loss": 0.6538,
       "step": 2810
     },
     {
       "epoch": 3.2,
-      "grad_norm": 14.510526657104492,
+      "grad_norm": 12.203571319580078,
       "learning_rate": 2.912621359223301e-06,
-      "loss": 0.2805,
+      "loss": 0.2297,
       "step": 2820
     },
     {
       "epoch": 3.2,
-      "grad_norm": 12.12006950378418,
+      "grad_norm": 10.11854362487793,
       "learning_rate": 2.725914861837192e-06,
-      "loss": 0.4065,
+      "loss": 0.4069,
       "step": 2830
     },
     {
       "epoch": 3.2,
-      "grad_norm": 5.202164173126221,
+      "grad_norm": 9.628511428833008,
       "learning_rate": 2.539208364451083e-06,
-      "loss": 0.3511,
+      "loss": 0.4468,
       "step": 2840
     },
     {
       "epoch": 3.21,
-      "grad_norm": 13.489916801452637,
+      "grad_norm": 13.15705394744873,
       "learning_rate": 2.352501867064974e-06,
-      "loss": 0.3357,
+      "loss": 0.3533,
       "step": 2850
     },
     {
       "epoch": 3.21,
-      "grad_norm": 12.779403686523438,
+      "grad_norm": 13.94404411315918,
       "learning_rate": 2.165795369678865e-06,
-      "loss": 0.367,
+      "loss": 0.3798,
       "step": 2860
     },
     {
       "epoch": 3.21,
-      "grad_norm": 13.837797164916992,
+      "grad_norm": 10.802663803100586,
       "learning_rate": 1.979088872292756e-06,
-      "loss": 0.4565,
+      "loss": 0.4173,
       "step": 2870
     },
     {
       "epoch": 3.22,
-      "grad_norm": 4.743872165679932,
+      "grad_norm": 6.215898513793945,
       "learning_rate": 1.7923823749066467e-06,
-      "loss": 0.3372,
+      "loss": 0.3416,
       "step": 2880
     },
     {
       "epoch": 3.22,
-      "grad_norm": 8.704170227050781,
+      "grad_norm": 8.513174057006836,
       "learning_rate": 1.6056758775205377e-06,
-      "loss": 0.3423,
+      "loss": 0.3657,
       "step": 2890
     },
     {
       "epoch": 3.22,
-      "grad_norm": 2.67067551612854,
+      "grad_norm": 1.2327791452407837,
       "learning_rate": 1.4189693801344288e-06,
-      "loss": 0.3444,
+      "loss": 0.3455,
       "step": 2900
     },
     {
       "epoch": 3.23,
-      "grad_norm": 11.893070220947266,
+      "grad_norm": 11.529572486877441,
       "learning_rate": 1.2322628827483198e-06,
-      "loss": 0.2581,
+      "loss": 0.2163,
       "step": 2910
     },
     {
       "epoch": 3.23,
-      "grad_norm": 4.046337127685547,
+      "grad_norm": 3.785139799118042,
       "learning_rate": 1.0455563853622106e-06,
-      "loss": 0.3214,
+      "loss": 0.3368,
       "step": 2920
     },
     {
       "epoch": 3.23,
-      "grad_norm": 11.184314727783203,
+      "grad_norm": 10.868538856506348,
       "learning_rate": 8.588498879761017e-07,
-      "loss": 0.3076,
+      "loss": 0.3235,
       "step": 2930
     },
     {
       "epoch": 3.24,
-      "grad_norm": 12.178153991699219,
+      "grad_norm": 12.210295677185059,
       "learning_rate": 6.721433905899926e-07,
-      "loss": 0.2403,
+      "loss": 0.2774,
       "step": 2940
     },
     {
       "epoch": 3.24,
-      "grad_norm": 1.377287745475769,
+      "grad_norm": 2.8030104637145996,
       "learning_rate": 4.854368932038835e-07,
-      "loss": 0.2376,
+      "loss": 0.2727,
       "step": 2950
     },
     {
       "epoch": 3.24,
-      "grad_norm": 14.243566513061523,
+      "grad_norm": 13.06821060180664,
       "learning_rate": 2.987303958177745e-07,
-      "loss": 0.3316,
+      "loss": 0.3094,
       "step": 2960
     },
     {
       "epoch": 3.25,
-      "grad_norm": 4.011444091796875,
+      "grad_norm": 6.2003912925720215,
       "learning_rate": 1.1202389843166542e-07,
-      "loss": 0.3262,
+      "loss": 0.3316,
       "step": 2970
     },
     {
       "epoch": 3.25,
-      "eval_accuracy": 0.7328859060402685,
-      "eval_loss": 0.7344102263450623,
-      "eval_runtime": 1687.1353,
-      "eval_samples_per_second": 0.442,
-      "eval_steps_per_second": 0.056,
+      "eval_accuracy": 0.72751677852349,
+      "eval_loss": 0.7167822122573853,
+      "eval_runtime": 1874.7039,
+      "eval_samples_per_second": 0.397,
+      "eval_steps_per_second": 0.05,
       "step": 2976
     },
     {
       "epoch": 3.25,
       "step": 2976,
       "total_flos": 6.0945535892058145e+19,
-      "train_loss": 0.7349759022234589,
-      "train_runtime": 150955.4956,
-      "train_samples_per_second": 0.158,
-      "train_steps_per_second": 0.02
+      "train_loss": 0.7312911278618279,
+      "train_runtime": 121527.2742,
+      "train_samples_per_second": 0.196,
+      "train_steps_per_second": 0.024
     },
     {
       "epoch": 3.25,
-      "eval_accuracy": 0.7043010752688172,
-      "eval_loss": 0.8327615261077881,
-      "eval_runtime": 1693.8047,
-      "eval_samples_per_second": 0.439,
-      "eval_steps_per_second": 0.055,
+      "eval_accuracy": 0.7163978494623656,
+      "eval_loss": 0.7914323806762695,
+      "eval_runtime": 1840.54,
+      "eval_samples_per_second": 0.404,
+      "eval_steps_per_second": 0.051,
       "step": 2976
     },
     {
       "epoch": 3.25,
-      "eval_accuracy": 0.7043010752688172,
-      "eval_loss": 0.8327615261077881,
-      "eval_runtime": 1683.6481,
-      "eval_samples_per_second": 0.442,
-      "eval_steps_per_second": 0.055,
+      "eval_accuracy": 0.7163978494623656,
+      "eval_loss": 0.7914324402809143,
+      "eval_runtime": 1837.684,
+      "eval_samples_per_second": 0.405,
+      "eval_steps_per_second": 0.051,
       "step": 2976
     }
   ],